Benchmark for evaluating relationship hallucinations in LVLMs. Features image-level and instance-level questions focusing on inter-object relationships.

	Fields

Field Description

image Input image

question_id Question identifier

question Question about image relationships

label Ground truth answer

qtype Question type

level Question level (image/instance)

subject Subject object

sub_box Subject bounding box

object Target object

obj_box… See the full description on the dataset page: https://huggingface.co/datasets/MM-Hallu/R-Bench.